MIF 1.0.0 is released

[zircote]

The Modeled Information Format (MIF) reached 1.0.0 today. This is the first stable release of the specification, the canonical JSON Schemas and the conversion tooling. The specification is published at https://mif-spec.dev.

MIF defines a portable format for AI memory, knowledge and information: a directory of Markdown concept files with YAML frontmatter, plus a derived JSON-LD projection for machines. Markdown is canonical. The JSON-LD form is regenerated from it and round-trips losslessly, so authors edit readable files while machine consumers receive linked data.

What went into 1.0.0

• A concept model built on the cognitive-psychology triad (semantic, episodic and procedural) as the base taxonomy, with a three-tier trait system for concepts that span more than one base type.
• Canonical JSON Schemas (JSON Schema 2020-12) published under stable, resolvable $id values on mif-spec.dev. Versioned mirrors (1.0.0, latest, v1) are generated at build time alongside an index catalog.
• Three conformance levels, gated in continuous integration by an Open Knowledge Format (OKF) conformance check and a lossless Markdown-to-JSON-LD round-trip check.
• The namespace vocabulary now resolves at https://mif-spec.dev/ns/, both as a human-readable reference and as a machine-readable RDFS document at https://mif-spec.dev/ns/vocabulary.jsonld. Both are derived from the canonical JSON-LD context, so the two representations cannot drift apart.
• The AI Memory profile, industry ontology examples and the mif_convert conversion tooling.
• Twelve Architecture Decision Records (ADRs) that document the reasoning, including the repositioning to "Modeled Information Format" and the relocation of ecosystem documentation to the dedicated doc-site repository.

Supply-chain attestation

Every 1.0.0 release artifact carries a cryptographic attestation. This attestation pipeline is new in this release: the release workflow was onboarded onto the organization's attested-delivery architecture during the cutover.

The workflow builds the source tarball and the schema bundle, generates SLSA build provenance and a CycloneDX Software Bill of Materials (SBOM) and verifies its own output before publication. The verification is fail-closed, which means a release that does not verify does not publish.

Anyone can validate an artifact independently:

gh release download v1.0.0 --repo modeled-information-format/MIF --pattern 'mif-1.0.0.tar.gz'
gh attestation verify mif-1.0.0.tar.gz --repo modeled-information-format/MIF

I validated both published artifacts (mif-1.0.0.tar.gz and mif-schemas-1.0.0.tar.gz) against their attestations after the release. Both verify against the GitHub Actions OIDC issuer, and the provenance names the exact workflow, commit and run that produced them. The validation confirms that what consumers download is byte-identical to what the pipeline built, signed and recorded.

A note of thanks

My thanks to @perlowja, whose input and contributions shaped the v1.0.0 schema additions. The optional W3C-PROV provenance layer (#85) and the vendor-neutral DocumentReference design (#84) came from that work, and both ship in this release. The motivation to carry a long-running migration across the line and turn it into a finished release came from the same source. Inspiration matters, and that one mattered here.

The specification, schemas and verification commands are all live at https://mif-spec.dev. Feedback and questions are welcome in this discussion.